3 ms·
> which is only going to get worse as it "learns" from garbage outputted by other LLMs moving forward You seem to assume that autoregressive pretraining (and u
by red75prime 4mo ago
> which is only going to get worse as it "learns" from garbage outputted by other LLMs moving forward
You seem to assume that autoregressive pretraining (and unfiltered behavior cloning, maybe) are the only ways to improve LLM performance.
- junon 4mo agoNo. I'm just saying, there decades worth of more PHP discourse than there is swift discourse.